Why Do Household Portfolio Shares Rise in Wealth?∗
Jessica A. Wachter†and Motohiro Yogo‡
Abstract
In the cross-section of U.S. households, the portfolio share in risky assets rises
in wealth. The standard life-cycle model with power utility and non-tradable labor
income has the counterfactual implication that the portfolio share declines in wealth.
We develop a life-cycle model in which household utility depends on two types of
consumption goods, basic and luxury. The model predicts that the expenditure share
for basic goods declines in total consumption, and the variance of consumption growth
rises in the level of consumption. When calibrated to match these two predictions in
household consumption data, the model explains portfolio shares that rise in wealth.
JEL classification: D11; D12; G11
Keywords: Decreasing relative risk aversion; Engel curves; Life-cycle model; Nonhomo-
thetic preferences; Portfolio choice
First draft: September 8, 2006
This draft: March 1, 2007
∗For comments and discussions, we thank Marshall Blume, John Campbell, Urban Jermann, Jose-Vıctor Rıos-Rull, Amir Yaron, Stephen Zeldes, and seminar participants at Wharton. We thank IndraneelChakraborty for valuable research assistance. This paper is based upon work supported under a Rodney L.White Center research grant.
†Finance Department, The Wharton School, University of Pennsylvania, 3620 Locust Walk, Philadelphia,PA 19104 and NBER (e-mail: [email protected]).
‡Finance Department, The Wharton School, University of Pennsylvania, 3620 Locust Walk, Philadelphia,PA 19104 and NBER (e-mail: [email protected]).
1 Introduction
Surveys of household finances reveal a striking fact: the share of household wealth in risky
assets rises in wealth. While poorer households are less likely to participate in the stock
market, this fact alone does not account for the positive relationship between wealth and
the share of financial wealth invested in stocks (hereafter, the portfolio share). The portfolio
share rises in wealth even among stockholding households. While more educated house-
holds tend to have higher portfolio shares, the portfolio share rises in wealth even among
stockholding households with the same education.
The standard life-cycle model in which the household has power utility and non-tradable
labor income predicts that the portfolio share declines in financial wealth (Bodie, Merton,
and Samuelson 1992). This prediction follows from the fact that the present value of future
labor income acts as a non-tradable “bond” in the household’s portfolio. At any given level
of human capital, optimal portfolio allocation requires that the household initially allocate
all of its financial wealth to stocks, and allocate part of its financial wealth to bonds only at
higher levels of wealth.
This paper examines the role that nonhomothetic utility plays in explaining the observed
relationship between portfolio choice and wealth. We develop a life-cycle model in which
the household has additive utility over two types of consumption goods. The household’s
utility function has higher curvature over one of these goods than the other. The first good
is “basic” in the sense that the expenditure share for this good falls in total consumption.
The expenditure share for the second good, called “luxury”, rises in total consumption. We
calibrate and solve the model with a labor income process that is standard in the life-cycle
literature (Carroll and Samwick 1997, Gourinchas and Parker 2002). We then simulate an
economy of ex ante identical households who are subject to idiosyncratic income shocks.
The model has three testable predictions for the cross-section of household consumption and
portfolio choice.
The first prediction is that the expenditure share falls in total consumption for some goods
2
and rises for others. Using household consumption data from the Consumer Expenditure
Survey (CEX), we identify basic goods as those goods whose expenditure shares fall in total
consumption. We find a significant relationship between the basic expenditure share and
wealth in the cross-section of households. In the nonhomothetic model, households with
higher permanent income allocate a lower share of their total consumption to the basic
good. Therefore, our empirical findings imply significant differences in the utility curvature
between the two types of goods. We use the estimated basic expenditure share to guide our
calibration of the life-cycle model.
The second prediction is that the variance of consumption growth rises in the level of con-
sumption. In the nonhomothetic model, households with higher permanent income are less
risk averse, and consequently their consumption responds more to wealth shocks. Using the
CEX, we confirm that the variance of consumption growth rises in the level of consumption,
even among stockholding households.
The third prediction is that the portfolio share rises in financial wealth. Households with
higher permanent income are less risk averse and consequently allocate a higher fraction of
their wealth to stocks. As in the power utility model, the nonhomothetic model also implies
that the optimal portfolio share declines in wealth, holding constant the level of permanent
income. However, such transitory variation in wealth is a less important determinant of
portfolio choice than the cross-sectional variation in permanent income, leading to a posi-
tive relationship between the portfolio share and wealth. We confirm this prediction with
household portfolio data from the Survey of Consumer Finances (SCF).1
This paper builds on an active literature on life-cycle models of consumption and port-
folio choice, calibrated to realistic processes for labor income.2 One branch of this literature
1The fact that the portfolio share in risky assets rises in wealth has been documented extensively invarious household surveys. Early evidence can be found in the 1962 and 1963 Surveys of the FinancialCharacteristics of Consumers and Changes in Family Finances (Blume and Friend 1975, Friend and Blume1975). Guiso, Haliassos, and Jappelli (2002, Table I.7) contains a summary of the international evidence forfive countries.
2See Bertaut and Haliassos (1997), Cocco, Gomes, and Maenhout (2005), Davis and Willen (2002),Gakidis (1998), Heaton and Lucas (1997), and Viceira (2001).
3
explains non-participation in the stock market through fixed costs of participation. Cocco
(2005), Hu (2005), and Yao and Zhang (2005) find that housing crowds out stocks in the
household’s portfolio and can explain non-participation in the presence of fixed costs. Gomes
and Michaelides (2005) find that low risk aversion, paired with a low elasticity of intertem-
poral substitution, leads to a low savings motive and can explain non-participation in the
presence of fixed costs. While non-participation is an important puzzle, we instead focus on
a different puzzle that the portfolio share rises in wealth conditional on participation, for
which there has been relatively little work.
Our work is also related to that of Carroll (2000, 2002), who proposes nonhomothetic
preferences in which wealth at the end of life is a luxury good. Carroll’s model has the
potential to explain the high savings rate and the portfolio behavior of the very wealthy,
at the top one percentile of the wealth distribution.3 Nonhomothetic preferences also play
an important role in the work of Aıt-Sahalia, Parker, and Yogo (2004), who develop a
representative household model with utility over two goods. They find that the consumption
of luxury goods, constructed from data on the sales of luxury retailers, is consistent with the
high historical equity premium. Unlike these previous studies, this paper calibrates and solves
a life-cycle model to generate quantitatively testable implications for the cross-sectional
distribution of consumption and portfolio choice. In addition, while these previous studies
have focused on households at the very peak of the wealth distribution, our work attempts
to explain consumption and portfolio choice across the entire distribution of stockholding
households.
The remainder of the paper proceeds as follows. Section 2 documents relevant facts about
household consumption, and Section 3 documents relevant facts about household portfolio
choice. Section 4 develops a life-cycle model with nonhomothetic preferences and describes
the preference and income parameters used in the calibration. Section 5 solves the model and
describes the policy functions for consumption and portfolio choice. Section 6 compares the
3In related work, Roussanov (2006) develops a model in which the household’s social status is effectivelya luxury good, which explains why wealthier households own portfolios with undiversified private equity.
4
consumption and portfolio behavior of simulated households to that of actual households
in household data. Section 7 concludes. The appendices contain further details on the
household data and the numerical methods used in solving the life-cycle model.
2 Facts about Household Consumption
In this section, we document relevant facts about the consumption behavior of U.S. house-
holds. Data on household consumption are from the CEX; a detailed description of the data
is given in Appendix A. Our unit of analysis is annual consumption so that each household
accounts for one observation in the dataset. We focus on the consumption of nondurable
goods and services for the following major categories of expenditure.
• Nondurable goods: Food at home; food away from home; clothing and shoes; fuel oil,
coal, and gasoline; and other nondurable goods.
• Services: Household operation; transportation; personal care; personal business; and
recreation.
Whenever we compare the level of consumption across households, we control for house-
hold characteristics, using a procedure similar to that in Carroll and Samwick (1997) and
Gourinchas and Parker (2002). We regress log consumption on a set of dummy variables for
marital status, household size, and birth cohort. For each household, we use the estimated
coefficients to impute the equivalent consumption for a household that is married, that has
four members, and whose head is born in the 1945–1949 cohort.4
We document household consumption behavior by age and total consumption. We first
sort households into age groups of ten years, according to the age of the household head.
Within each age group, we then sort households into quartiles by their total consumption.
We sort households by consumption, rather than financial wealth, because consumption data
4This approach implicitly assumes that there are cohort but not time effects. We have repeated theexercise assuming time but not cohort effects and found very similar results.
5
are more complete and reliable than financial data in the CEX. However, we have verified
that our main findings are robust to sorting households by financial wealth. For each age
group, we create an additional bin for households whose consumption is in the top five
percentile, in order to separately analyze the behavior of the wealthy. Panel A of Table 1
reports the median consumption within each bin. The last row of the panel reports results
for all households in that age group.
2.1 Basic versus Luxury Consumption
We first document facts about expenditure shares by age and total consumption. Table 2
reports the median expenditure share for all the components of nondurable consumption.
The bins in the table are defined according to age group and total consumption quartile,
reported in Panel A of Table 1. As shown in Panel A, food at home as a share of total
consumption declines in total consumption within each age group. For the 36–45 age group,
food at home is 31.2% of total consumption for the lowest consumption quartile and 17.3%
for the highest quartile. For the 56–65 age group, food at home is 31.3% of total consumption
for the lowest consumption quartile and 16.8% for the highest quartile.
In contrast, as shown in Panel B, food away from home as a share of total consumption
rises in total consumption within each age group. For the 36–45 age group, food away from
home is 4.5% of total consumption for the lowest consumption quartile and 9.7% for the
highest quartile. For the 56–65 age group, food away from home is 2.8% of total consumption
for the lowest consumption quartile and 9.7% for the highest quartile. The fact that the
expenditure share rises in total consumption is consistent with introspection, which suggests
that dining out frequently is a luxury that mainly wealthier households can afford.
As shown in Panel C, clothing and shoes as a share of total consumption rises in total
consumption within each age group. As shown in Panel D, fuel oil, coal, and gasoline as a
share of total consumption falls in total consumption within each age group. As shown in
Panel E, other nondurable goods as a share of total consumption rises in total consumption
6
within each age group, with the exception of the highest consumption quartile.
Table 3 reports the median expenditure share for all the components of service consump-
tion. The bins in the table are defined according to age group and total consumption quartile,
reported in Panel A of Table 1. As shown in Panel A, household operation as a share of
total consumption falls in total consumption within each age group. As shown in Panel B,
transportation as a share of total consumption rises in total consumption within each age
group. As shown in Panel C, personal care as a share of total consumption rises in total
consumption within each age group. As shown in Panel D, personal business (accounting,
tax services, banking) as a share of total consumption rises in total consumption within each
age group. As shown in Panel E, recreation as a share of total consumption rises in total
consumption within each age group.
Tables 2 and 3 map out the household Engel curves, that is, the variation in expenditure
shares as a function of total consumption. Our working definition of basic goods is those
nondurable goods and services whose expenditure shares fall in total consumption. Luxury
goods are those nondurable goods and services whose expenditure shares rise in total con-
sumption. Based on these definitions, we construct basic and luxury consumption as the
sum of the following categories of expenditure.
• Basic consumption: Food at home; fuel oil, coal, and gasoline; and household operation.
• Luxury consumption: Food away from home; clothing and shoes; other nondurable
goods; transportation; personal care; personal business; and recreation.
Panel B of Table 1 reports basic consumption as a share of total consumption. For the
36–45 age group, basic consumption is 64.0% of total consumption for the lowest consump-
tion quartile and 42.1% for the highest quartile. The basic expenditure share is 35.4% for
households in the top five percentile of total consumption. For the 56–65 age group, basic
consumption is 68.1% of total consumption for the lowest consumption quartile and 41.7%
for the highest quartile. The basic expenditure share is 31.1% for households in the top five
7
percentile of total consumption.
2.2 Consumption Volatility
We now document facts about consumption volatility by age and total consumption. Es-
timation of the variance of consumption growth is complicated by the fact that household
consumption data are subject to considerable measurement error. Because measurement er-
ror is (by definition) transitory, we cannot identify the variance of the transitory component
of consumption. However, we can identify the variance of the permanent component of con-
sumption, following an identification strategy similar to that used by Carroll and Samwick
(1997) to identify the variance of the permanent component of income.
Each household in the CEX is interviewed for four consecutive quarters. Let C1, . . . , C4
denote quarterly consumption in interviews one through four. For each household, we com-
pute a statistic that we refer to as consumption volatility,
σ2C =
1
2
[(log
C3
C1
)2
+
(log
C4
C2
)2]− 1
3
[(log
C2
C1
)2
+
(log
C3
C2
)2
+
(log
C4
C3
)2]
. (1)
Suppose log consumption can be decomposed into permanent and transitory components as
logCt+T
Ct
=T∑
s=1
ηt+s + εt+T − εt. (2)
For mean zero iid shocks εt, εt+T . Then the variance of the permanent component of log
consumption can be identified through the moment condition
E[σ2C ] = σ2
η, (3)
where the expectation is taken over the cross-section of households.
Panel C of Table 1 reports the square root of the median of consumption volatility
8
within each bin, defined according to age group and total consumption quartile.5 Although
consumption volatility can be negative for a given household, it is positive for the median
household. Consumption volatility rises in total consumption within each age group, as
documented by Yogo (2006). For the 36–45 age group, consumption volatility is 9.2% for
the lowest consumption quartile and 10.2% for the highest quartile. Consumption volatility
is 15.8% for households in the top five percentile of total consumption. For the 56–65 age
group, consumption volatility is 10.2% for the lowest consumption quartile and 11.4% for the
highest quartile. Consumption volatility is 15.8% for households in the top five percentile of
total consumption.
As documented by Mankiw and Zeldes (1991), households that own stock have consump-
tion that is more volatile than those that do not. To the extent that stockholding is positively
related to wealth, a relevant question is whether the relationship in Panel C is explained (at
least partly) by stockholding. As shown in Table 9 below, however, consumption volatility
rises in total consumption even within the sub-sample of stockholding households. In fact,
the relationship is even stronger for stockholding households.
3 Facts about Household Portfolios
In this section, we document relevant facts about the portfolio behavior of U.S. households.
Data on household portfolios are from the SCF; a detailed description of the data is given in
Appendix B. The SCF is a repeated cross-section of households every three years, available
in the current format for the period 1989–2004. The main results that we report here are
for the 2001 SCF. However, we find qualitatively similar results for the other years of the
survey, which are reported in Wachter and Yogo (2007).
Our working definition of the portfolio share is the share of financial wealth invested in
stocks. A broader definition that has also been employed in the literature is the share of
net worth invested in risky assets, which includes corporate, foreign, and mortgage-backed
5We report the median rather than the mean because it is less sensitive to measurement error.
9
bonds; business equity; and investment real estate (see Bertaut and Starr-McCluer (2002,
Table 5.7)). In Wachter and Yogo (2007), we find qualitatively similar results for this alter-
native definition of the portfolio share.
We document household portfolio behavior by age and financial wealth. As emphasized
by Ameriks and Zeldes (2004), the fact that age, time, and cohort effects are not separately
identified complicates the interpretation of household portfolio behavior. We focus on the
2001 SCF and control for age. The implicit assumption in our analysis is that there is
no cohort effect, which is the standard practice in the literature (e.g., Bertaut and Starr-
McCluer (2002) and Campbell (2006)). We therefore abstain from interpreting age patterns
in portfolio behavior, which could arise from a cohort effect. The focus of our analysis will
be the relationship between financial wealth and the portfolio share, controlling for age.
3.1 Portfolio Share by Wealth
We first sort households into age groups of ten years, according to the age of the household
head. Within each age group, we then sort households into quartiles by their financial wealth.
For each age group, we create an additional bin for households whose financial wealth is in
the top five percentile, in order to separately analyze the behavior of the wealthy. Panel A
of Table 4 reports the median financial wealth within each bin. The last row of the panel
reports results for all households in that age group.
Panel B reports the percentage of households that hold stocks within each age group
and wealth quartile. The average participation rate is well below 100%, as documented
by Haliassos and Bertaut (1995) and Mankiw and Zeldes (1991). For each age group, the
participation rate rises rapidly in financial wealth. For the middle three age groups, the
participation rate is above 90% in the highest wealth quartile. In contrast, for all age
groups, the participation rate is below 10% in the lowest wealth quartile.
Panel C reports the median portfolio share in stocks within each age group and wealth
quartile. Given the low participation rates reported in Panel B, it is unsurprising that
10
the median portfolio share is quite low for households in the lower wealth quartiles. Like
participation, the median portfolio share also rises rapidly in financial wealth, with values
ranging from 45% to 60% in the top wealth quartile.6
Table 4 reveals an age profile in portfolio choice that is consistent with those reported
in Ameriks and Zeldes (2004). In Panel B, the participation rate first rises and then falls
in age. This hump-shaped pattern is also evident within each of the wealth quartiles. In
Panel C, the portfolio share for all households also inherits the hump-shaped pattern in age.
3.2 Portfolio Share for Stockholders
In Table 5, we restrict the sample to only those households that own stock. We again sort
households by age, and then into quartiles of financial wealth within each age group. Panel A
reports the median financial wealth within each age group and wealth quartile. The level of
financial wealth in each bin is higher than those in Panel A of Table 4 because stockholders
tend to be wealthier than nonstockholders.
Panel B of Table 5 reports the median portfolio share for stockholders. This panel
shows that the relationship between financial wealth and the portfolio share is not due to
participation alone. For the 36–45 age group, the portfolio share is 43% for the lowest wealth
quartile and 68% for the highest quartile. The portfolio share is 69% for households in the
top five percentile of financial wealth. For the 56–65 age group, the portfolio share is 45%
for the lowest wealth quartile and 54% for the highest quartile. The portfolio share is 55%
for households in the top five percentile of financial wealth.
6This relationship between wealth and portfolio share and the refinements that follow may seem at oddswith recent findings by Brunnermeier and Nagel (2005). A way to reconcile these studies is that householdportfolios do not respond to changes in wealth at higher frequencies, perhaps due to various transactioncosts, but they respond at a much lower life-cycle frequency.
11
3.3 Portfolio Share by Education and Wealth
More educated households tend to be wealthier, and more educated investors tend to have a
higher share of their financial assets invested risky assets (see Campbell (2006)). Therefore,
the relationship between financial wealth and the portfolio share may be explained (at least
partly) by education. To address this issue, Table 6 repeats the analysis in Table 5, separately
by education group. The three education groups are households whose head is a high school
graduate, whose head has some college education, or whose head is a college graduate. For
each education group, we again sort households by age, and then into quartiles of financial
wealth within each age group. Even controlling for eduction, the portfolio share rises in
financial wealth, and the effect is generally as strong as that found in Table 5.7
Our tabulations allow the portfolio share to depend on age, wealth, stockholding, and ed-
ucation in potentially nonlinear ways. We have shown that the positive relationship between
financial wealth and the portfolio share is robust to these covariates. However, many factors
other than age and education can influence household portfolio choice. Proper control for a
large number of factors is most easily implemented through a rich regression model. Esti-
mates of such models indicate that the portfolio share rises in wealth, even in the presence of
an extensive set of controls (see Bertaut and Starr-McCluer (2002) for the SCF and Calvet,
Campbell, and Sodini (2006) for a Swedish dataset).
7Among older, wealthier households, the median portfolio share falls off slightly from the third to thefourth wealth quartile. This pattern can be explained by the increased ownership in private equity amongthis group, which crowds out holding of other risky assets (Heaton and Lucas 2000). See Wachter and Yogo(2007) for tabulations of the risky-asset share, which includes private equity.
12
4 A Life-Cycle Model of Consumption and Portfolio
Choice
4.1 Nonhomothetic Utility
We assume that the household consumes two types of nondurable goods, denoted B for basic
and L for luxury. The household has additively separable and nonhomothetic utility over
the two goods, specified as
U(B,L) =B1−γ
1 − γ+ α
L1−φ
1 − φ. (4)
The curvature for the basic good is higher than that for the luxury good, that is γ > φ > 0.
The parameter α ≥ 0 is the utility weight on the luxury good. This utility specification can
be generalized to a model with multiple goods and multiple corresponding curvatures. The
two-good model can be thought of as a simplification of such a model that captures the main
economic effects.
For simplicity, we assume that the relative price between the two goods is constant and
normalize it to be one. Optimal consumption of the two goods requires that
B−γ = αL−φ. (5)
Let C = B + L denote the total consumption expenditure. The first-order condition implies
the expenditure shares
B
C=
1
1 + α1/φBγ/φ−1, (6)
L
C=
1
1 + α−1/γLφ/γ−1. (7)
The expenditure share for the basic good falls, and the expenditure share for the luxury
13
good rises in total consumption.8
An important implication of nonhomothetic utility is decreasing relative risk aversion
(Stiglitz 1969). By Hanoch (1977, Theorem 1), the household’s relative risk aversion is given
by
RRA =
(γ−1 B
C+ φ−1 L
C
)−1
. (8)
When the household is poor, it consumes mostly basics, and its relative risk aversion is close
to γ. As the household grows wealthier, it consumes more luxuries, and its relative risk
aversion falls toward φ. The elasticity of intertemporal substitution is the reciprocal of its
relative risk aversion. The household has a higher elasticity of intertemporal substitution for
the luxury good than it does for the basic good (Browning and Crossley 2000).
To highlight the distinct implications of the nonhomothetic model, we also solve the
standard model with power utility
U(C) =C1−γ
1 − γ. (9)
When utility is homothetic across goods, the household’s indirect utility can be specified as
power utility over total consumption. Therefore, power utility can be thought of as a special
case of nonhomothetic utility (4) when γ = φ. The power utility model implies constant
relative risk aversion.
4.2 Life-Cycle Problem
We solve a life-cycle consumption and portfolio-choice problem for a household with nonho-
mothetic utility (4). The household starts adult life with initial financial wealth W1. The
household enters each period t = 1, . . . , T with cash-on-hand Wt, which is composed of finan-
8Our utility specification implies that the household consumes a positive amount of both goods. Anegative subsistence level for the luxury good would allow the household to consume none of the luxurygood at low levels of total consumption (see Aıt-Sahalia, Parker, and Yogo (2004)).
14
cial assets and labor income Yt. The household’s total consumption is Ct, which is optimally
allocated between basics Bt and luxuries Lt. Financial wealth remaining after consumption,
St = Wt −Ct, is saved either in bonds or stocks. Bonds have a constant gross rate of return
Rf , and stocks have a stochastic gross rate of return Ret. The household is subject to a bor-
rowing constraint, so that St ≥ 0. The household is also subject to a short-sales constraint,
so that the portfolio share in stocks must satisfy at ∈ [0, 1].
Let β ∈ (0, 1) denote the household’s subjective discount factor. The household’s problem
is to choose consumption and the portfolio share in each period to maximize the expected
discounted sum of future utility flow,
E1
T∑t=1
βt−1U(Bt, Lt). (10)
The household is subject to the intertemporal budget constraint
Wt+1 = Rt+1(Wt − Bt − Lt) + Yt+1, (11)
Rt+1 = Rf + at(Re,t+1 − Rf ). (12)
4.3 Calibration of the Model
Following Carroll (1997), we calibrate the model to a 50-year life cycle. The household works
and earns labor income from ages 26 through 65 (t = 1, . . . , 40). The household is retired
from ages 66 through 75 (t = 41, . . . , 50), and then subsequently dies. Table 7 summarizes
the parameters in the benchmark calibration, which we now discuss in more detail.
4.3.1 Preferences
For the power utility model, we set the preference parameters to β = 0.96 and γ = 5, which
are standard in the literature (e.g., Cocco, Gomes, and Maenhout (2005) and Gomes and
Michaelides (2005)).
15
For the nonhomothetic model, we choose the same discount factor β = 0.96. We set
the curvature parameters to γ = 7 and φ = 3, which bracket the value used for calibrating
the power utility model. The magnitude of the curvature parameters determines relative
risk aversion (8) and has implications for consumption volatility and portfolio allocation.
As revealed by equation (6), the relative magnitude of the curvature parameters determines
how the basic expenditure share varies in total consumption. Therefore, the relative cur-
vature γ/φ is chosen to lead to realistic variation in the basic expenditure share across the
consumption quartiles. Given the curvature parameters, we set α = 10 to match the median
basic expenditure share for stockholders in the CEX (see Table 8).
4.3.2 Labor Income
Following Carroll (1997) and Zeldes (1989), we model the household’s stochastic labor income
as
Yt = Ptεt, (13)
Pt+1 = Gt+1Ptηt+1, (14)
given an initial level P0. The variable Pt denotes “permanent income” in period t, defined
as the labor income that would be earned in the absence of transitory shocks (i.e., εt = 1).
Permanent income has a deterministic component that grows at the rate Gt each period.
During the household’s working life (through age 65), permanent income is subject to an
independently and identically distributed (i.i.d.) shock
log ηt ∼ N(−σ2η/2, σ
2η), (15)
where N denotes the normal distribution. The permanent shock satisfies E[ηt] = 1.
16
During the working life, labor income is also subject to an i.i.d. transitory shock
εt =
⎧⎪⎨⎪⎩ 0 with probability p
εt with probability 1 − p, (16)
log εt ∼ N(µε, σ2ε ). (17)
Unemployment occurs with probability p. The parameter µε is chosen so that the transitory
shock satisfies E[εt] = 1. In our benchmark case, we set the probability of unemployment to
zero and therefore set µε = −σ2ε /2 (as in Cocco, Gomes, and Maenhout (2005) and Gomes
and Michaelides (2005)). We also consider a case that allows for a positive probability of
unemployment. During retirement, the household receives nonstochastic income that grows
at the rate Gt.
We calibrate the labor income process using standard parameters in the life-cycle con-
sumption literature (Carroll and Samwick 1997, Gourinchas and Parker 2002). The variance
of the income shocks are set to σ2η = 2.12% and σ2
ε = 4.40%. In the calibration with
unemployment, the probability of unemployment in any given year is set to 0.5% (Carroll
1992).
In order to calibrate the deterministic component of income, we follow Gourinchas and
Parker (2002) and estimate average life-cycle income using CEX data on disposable income.
We regress log disposable income on a third degree polynomial in age, which is interacted
with a dummy variable for whether or not the household is retired. The regression also
includes dummy variables for marital status, household size, and birth cohort. We use the
estimated coefficients to build the life-cycle income profile for a “typical” household that
works from ages 26 through 65 and is retired from ages 66 through 75 (at the retirement
date, labor income is estimated to fall by 23.5% relative to the previous period). We calibrate
the growth rate to that of a household that is married, that has four members, and whose
head is born in the 1945–1949 cohort. Further details are contained in Wachter and Yogo
(2007).
17
4.3.3 Asset Returns
We calibrate asset returns using a standard specification in the life-cycle portfolio-choice
literature (e.g., Gomes and Michaelides (2005)). The bond return is set to 2% per year, and
the equity premium is set to 4% per year. Stock returns are distributed as
Ret = Reνt, (18)
log νt ∼ N(−σ2ν/2, σ
2ν). (19)
The shock to stock returns satisfies E[νt] = 1, and its variance is set to σν = 18%.
The correlation between the shocks to stock returns and permanent income is set to ρ =
E[νtηt]/(σνση) = 15% as estimated in the prior literature (see Campbell, Cocco, Gomes, and
Maenhout (2001) and Gomes and Michaelides (2005)).
4.4 Discussion of the Model
In modeling household consumption and portfolio choice, we have made several simplifying
assumptions. The assumptions allow us to focus on the portfolio implications of the non-
homothetic model in the simplest setting. We now discuss these assumptions briefly and
provide some intuition for how modifications of these asssumptions are likely to affect our
results.
We assume that the relative price of luxury goods is constant. A time trend in the
relative price of goods can induce a time trend in the relative expenditure shares of those
goods. Instead of introducing a time trend in the relative price through the model, we take
out potential time effects in the data. Our tabulations of the basic expenditure share are
conditional on age and cohort, so that we have effectively controlled for time effects. We
have also repeated the same tabulations controlling for age and time effects with essentially
the same results. Even in the absence of a time trend, the relative price could affect portfolio
choice if shocks to the relative price are correlated with stock returns. Indeed, Aıt-Sahalia,
18
Parker, and Yogo (2004) find that growth rate of the price of luxury goods is highly positively
correlated with stock returns. In the presence of such correlation, a wealthy household, who
has a higher expenditure share for luxury goods, has an incentive to holds stocks to partially
hedge the price risk of luxury goods. Therefore, such correlation can magnify the positive
relationship between wealth and the portfolio share.
We assume an identical income process for all households. In order for income hetero-
geneity to explain the empirical patterns in consumption volatility (Table 1), it would be
necessary for the volatility of labor income to be higher for wealthier households. However,
more volatile labor income would lead to lower portfolio shares for wealthier households,
making the evidence in Table 5 more difficult to explain. Moreover, Carroll and Samwick
(1997, Table 1) report the variance of permanent income shocks by education: 2.1% for
some high school education, 2.8% for high school graduates, 2.4% for some college educa-
tion, 1.5% for college graduates, and 1.2% for graduate school. Insofar as education proxies
for permanent income, the variance of permanent income shocks appears to fall in the level
of permanent income. To verify that income heterogeneity is not driving our results, we
calibrate the model separately by education group and report the results in Wachter and
Yogo (2007).
We assume a constant investment opportunity set, that is, expected stock returns are
constant. A time-varying opportunity set generates life-cycle patterns in stock ownership,
but is unlikely to affect the relationship between wealth and the portfolio share that is the
focus of this paper (see Balduzzi and Lynch (1999), Barberis (2000), Campbell and Viceira
(1999), Kim and Omberg (1996), and Wachter (2002)).
5 Solution of the Life-Cycle Model
We solve the life-cycle problem through numerical dynamic programming as described in
Appendix C. As shown in the appendix, the household’s value function can be written as
19
a function of age (t), permanent income (Pt), and normalized cash-on-hand (wt = Wt/Pt).
This section describes the optimal policies for consumption and portfolio choice.
5.1 Optimal Consumption Policy
Figure 1 shows the optimal consumption policy, as a function of normalized cash-on-hand and
permanent income, for the household at age 50. The policy variables are basic consumption
(bt = Bt/Pt) and luxury consumption (lt = Lt/Pt), both normalized by permanent income.
Holding fixed the level of permanent income, the consumption function for both basics and
luxuries share the two key features of the standard consumption function in the power utility
model. First, the consumption function is monotonic in cash-on-hand. The higher is current
wealth, the higher is the consumption of both basics and luxuries. Second, the consumption
function is concave in cash-on-hand. The consumption of both basics and luxuries rises
steeply at lower levels of cash-on-hand. The marginal propensity to consume (MPC) out of
wealth then levels off at higher levels of cash-on-hand.
Holding fixed the level of normalized cash-on-hand, bt falls and lt rises in permanent
income. Wealthier households allocate a higher share of total consumption to the luxury
good. This effect is a consequence of the nonlinear Engel curves, described in Section 4.1.
At a low level of permanent income, basic consumption rises sharply in cash-on-hand, while
luxury consumption is relatively flat in cash-on-hand. In other words, the MPC for basic
consumption is high, while the MPC for luxury consumption is low. The exact opposite holds
at a high level of permanent income. The MPC for luxury consumption rises more rapidly in
permanent income in comparison to the MPC for basic consumption. This implies that total
consumption is more responsive to cash-on-hand at higher levels of permanent income. This
effect translates to a testable implication, that wealthier households should have consumption
that is more volatile than poorer households.
20
5.2 Optimal Portfolio Policy
Figure 1 shows the optimal portfolio policy, as a function of normalized cash-on-hand and
permanent income, for the household at age 50. The policy variable is the portfolio share, the
percent of financial wealth allocated to stocks. Holding fixed the level of permanent income,
the portfolio share falls in cash-on-hand, just as it does in the power utility model. Because
labor income is relatively stable and has a low correlation with stock returns, human capital
is nearly a substitute for bonds. The lower is cash-on-hand relative to permanent income, the
lower are stock holdings as a share of total (financial and non-financial) wealth. Therefore,
the lower is normalized cash-on-hand, the higher is the optimal allocation to stocks as a
share of financial wealth.
At a low level of permanent income, a higher share of total consumption is allocated to
basics, and the household resembles a power utility investor with higher risk aversion γ. At a
high level of permanent income, a higher share of total consumption is allocated to luxuries,
and the household resembles a power utility investor with lower risk aversion φ. Therefore,
holding fixed the level of normalized cash-on-hand, the household allocates a higher share of
financial wealth to stocks at higher levels of permanent income.
6 Simulation of the Life-Cycle Model
In order to assess the quantitative implications of the model, we simulate a cross-section of
10,000 households at an annual frequency. The households are ex ante identical, have non-
homothetic utility, and face non-tradable labor income, with the parameters summarized in
Table 7. For each household, we draw an initial level of financial wealth (relative to perma-
nent income) from a lognormal distribution, based on estimates from the CEX (Gourinchas
and Parker 2002, Table 2). The mean of W1/P1 is set to 0.3, and its log standard deviation
is set to 1.784. Similarly, the initial level of permanent income is drawn from a lognormal
distribution, based on estimates from the CEX (see Wachter and Yogo (2007)). The mean of
21
P1 is normalized to one, and its log standard deviation is set to 0.562. In order to highlight
the novel implications of the nonhomothetic model, we repeat the same simulation exercise
for power utility.
In our benchmark model, it is always optimal for the household to own some equity,
regardless of the level of wealth. In reality, however, a significant fraction of the population
does not own equity, as documented in Table 4. Explanations for non-participation range
from fixed costs of participation to investor mistakes (Calvet, Campbell, and Sodini 2006).
Because non-participation is outside the scope of our study, we interpret our model as a
description of stockholding households (i.e., those that have already paid the fixed cost and
are making optimal decisions). In order to assure that the model’s moments are comparable
to the empirical moments, all of the CEX and SCF tabulations in this section are for the
sub-sample of stockholding households. Therefore, the empirical moments differ from those
discussed in Sections 2 and 3, which are for all households.
6.1 Implications for Consumption
6.1.1 Basic Expenditure Share
We first look at the implications of our model for consumption data. Panel A of Table 8
reports the basic expenditure share for stockholding households in the CEX, tabulated by
age group and consumption quartile. For the 36–45 age group, basic consumption is 56% of
total consumption for the lowest consumption quartile and 38% for the highest quartile. The
basic expenditure share is 31% for households in the top five percentile of total consumption.
For the 56–65 age group, basic consumption is 56% of total consumption for the lowest
consumption quartile and 36% for the highest quartile. The basic expenditure share is 26%
for households in the top five percentile of total consumption.
In Panel B, we sort households simulated in the nonhomothetic model by age, and then
consumption quartile within each age group. We then tabulate the median of basic expendi-
ture share within each bin and compare the results with the empirical moments in Panel A.
22
Within each age group, the basic expenditure share falls in total consumption, comparing
favorably to the empirical moments. For the 36–45 age group, basic consumption is 73%
of total consumption for the lowest consumption quartile and 38% for the highest quartile.
The basic expenditure share is 30% for households in the top five percentile of total con-
sumption. For the 56–65 age group, basic consumption is 68% of total consumption for the
lowest consumption quartile and 30% for the highest quartile. The basic expenditure share
is 23% for households in the top five percentile of total consumption.
These results can be explained by the shape of the policy functions in Figure 1. As house-
holds become wealthier, their consumption of luxuries (the good with the lower curvature)
rises relative to their consumption of basics (the good with the higher curvature).
6.1.2 Consumption Volatility
Panel A of Table 9 reports the consumption volatility for stockholding households in the
CEX, tabulated by age group and consumption quartile. As discussed in Section 2, the
CEX data show striking differences in consumption volatility between low and high wealth
households. Panel A shows that this relationship also holds in the sub-sample of stock-
holding households. For the 36–45 age group, consumption volatility is 7.6% for the lowest
consumption quartile and 10.6% for the highest quartile. Consumption volatility is 15.3%
for households in the top five percentile of total consumption. For the 56–65 age group,
consumption volatility is 6.5% for the lowest consumption quartile and 12.4% for the highest
quartile. Consumption volatility is 24.8% for households in the top five percentile of total
consumption.
In Panel B, we compute consumption volatility as squared consumption growth for each
household simulated in the nonhomothetic model. We then tabulate the square root of
median consumption volatility within each bin and compare with the empirical moments in
Panel A. Except in the youngest age group, the nonhomothetic model generates consumption
volatility that rises in total consumption. For the 36–45 age group, consumption volatility
23
is 7.6% for the lowest consumption quartile and 9.0% for the highest quartile. Consumption
volatility is 8.9% for households in the top five percentile of total consumption. For the
56–65 age group, consumption volatility is 4.9% for the lowest consumption quartile and
6.6% for the highest quartile. Consumption volatility is 6.7% for households in the top five
percentile of total consumption.
These results can be explained by the fact that households with higher permanent in-
come have lower risk aversion in the nonhomothetic model. As shown in Figure 1, the MPC
for basic consumption falls gradually, while the MPC for luxury consumption rises rapidly
in permanent income. Therefore, luxury consumption accounts for much of the rise in con-
sumption volatility as a function of permanent income. Simply put, consumption fluctuations
become more tolerable as the household becomes wealthier because luxury consumption is
more discretionary than basic consumption.
For the purposes of comparison, Panel C repeats the same exercise for the power utility
model and shows that consumption volatility is roughly constant within age group, which is
at odds with the data. Comparing Panels B and C, there are two features of consumption
volatility that are common to both the nonhomothetic and the power utility model. First,
consumption volatility is very high for low consumption households in the 26–35 age group.
Younger households have had less time to accumulate wealth and are less able to buffer
income shocks. Second, consumption volatility falls in age because younger households, who
have accumulated less wealth, have a higher MPC than do older households.
6.2 Implications for Portfolio Choice
Panel A of Table 10 reports the median portfolio share for stockholding households in the
SCF, tabulated by age group and wealth quartile. Panel B reports the portfolio share for
households simulated in the nonhomothetic model, and Panel C reports the portfolio share
in the power utility model.
In the nonhomothetic model, the portfolio share rises in financial wealth for all age groups,
24
consistent with the evidence in Panel A. For the 36–45 age group, the portfolio share is 68%
for the lowest wealth quartile and 79% for the highest quartile. The portfolio share is 82%
for households in the top five percentile of financial wealth. For the 56–65 age group, the
portfolio share is 40% for the lowest wealth quartile and 58% for the highest quartile. The
portfolio share is 59% for households in the top five percentile of financial wealth. The rise
in the portfolio share becomes more pronounced as households age.
In stark contrast, the portfolio share falls in wealth in the power utility model. This fall
in the portfolio share becomes less pronounced as households age because human capital
becomes a smaller fraction of the household’s portfolio. For older households, the model
more closely resembles that of Samuelson (1969), in which all wealth is financial and the
optimal portfolio share is constant in wealth.
The key to understanding the findings for the nonhomothetic model lies in the port-
folio policy function, shown in Figure 2. The optimal portfolio share falls in normalized
cash-on-hand and rises in permanent income. As a result, there are two offsetting effects
that determine the relationship between financial wealth and the portfolio share in the non-
homothetic model. On the one hand, households with higher financial wealth have higher
normalized cash-on-hand, holding constant the level of permanent income. This is the stan-
dard effect that is also present in the power utility model, which causes households with
higher wealth to hold more equity to diversify human capital. On the other hand, higher
financial wealth implies higher permanent income, holding constant the level of normalized
cash-on-hand. Higher permanent income leads to lower risk aversion under nonhomothetic
utility and consequently a higher allocation of financial assets to stocks. As households age
and permanent income shocks accumulate, the cross-sectional dispersion in permanent in-
come rises. This effect explains why the spread in the portfolio shares between high and low
wealth households becomes more pronounced in age.
25
6.2.1 Discussion of the Age Profile
In both models, the portfolio share falls in age, which is a standard implication of the life-cycle
model (see Cocco, Gomes, and Maenhout (2005)). Households are born with little financial
wealth, but a large stake in non-tradable human capital. Because stocks have a high average
rate of return and low correlation with labor income, optimal portfolio allocation requires
that households initially allocate most of their financial wealth to stocks. As households age,
they accumulate financial wealth and depreciate their human capital. Because there is less
need to diversify human capital, the portfolio share in stocks falls.
As shown in Panel A, the portfolio share in the SCF appears mostly flat in age. There
are several proposed explanations for the discrepancy between the standard implication of
the model and the data. First, the true relationship between age and the portfolio share is
unknown because of the lack of identification between age, time, and cohort effects (Ameriks
and Zeldes 2004). Second, the purchase of housing and small fixed costs can crowd out stocks
from the household’s portfolio early in life (Cocco 2005, Hu 2005, Yao and Zhang 2005).
Third, internal habit formation can induce a strong motive to save in the riskfree asset early
in life, crowding out stocks from the household’s portfolio (Gomes and Michaelides 2003,
Polkovnichenko 2007). Finally, different assumptions on the joint process for stock returns
and labor income can substantially reduce stockholding for younger households (Benzoni,
Collin-Dufresne, and Goldstein 2007, Lynch and Tan 2006, Storesletten, Telmer, and Yaron
2007).
Rather than taking a stance on these various explanations for age effects in household
portfolios, we have focused on the relationship between wealth and the portfolio share con-
ditional on age. We could simultaneously explain the age profile in portfolio choice by
incorporating one of these additional ingredients. We leave this issue for future research.
26
6.3 Portfolio Choice with Unemployment Risk
We now examine how unemployment risk, or a positive probability of zero income, affects
portfolio choice. This scenario is of interest because the optimal portfolio share can rise in
cash-on-hand at a sufficiently low level of wealth, even under standard power utility (see
Cocco, Gomes, and Maenhout (2005)). Significant correlation between stock returns and
labor income can have a similar effect to unemployment risk, that is, the portfolio share can
rise in cash-on-hand at a sufficiently low level of wealth (Lynch and Tan 2006). However,
this effect disappears in age and should be nonexistent for retired households with no labor
income. Therefore, it cannot explain the fact that the relationship between wealth and the
portfolio share persists in age.
The parameters of the model remain the same as those in Table 7, except that labor
income can be zero in any period of the working life with probability 0.5%. Because the
implications for consumption are nearly identical to those in the benchmark case with no
unemployment, we focus on implications for portfolio choice in this section.
Table 11 shows the portfolio share for the nonhomothetic model in Panel B and for the
power utility model in Panel C. The portfolio share by age group and wealth quartile from the
SCF are repeated in Panel A for the purposes of comparison to the models. Unemployment
risk lowers the allocation to equity for youngest households in the lowest wealth quartile.
However, it hardly affects older households who have accumulated enough wealth to buffer
these temporary shocks to labor income. Given the preference and income parameters that
have realistic implications for household behavior, unemployment risk does not have a large
effect on portfolio choice.
7 Conclusion
Evidence from the CEX shows that the expenditure share for various categories of non-
durable goods and services vary significantly in total consumption. Moreover, the variance
27
of consumption growth rises in the level of consumption. Evidence from the SCF shows
that the portfolio share in stocks rises in wealth, even after controlling for stock market
participation and education. All these facts from household data are inconsistent with the
standard life-cycle consumption and portfolio choice model with power utility.
In this paper, we propose a parsimonious model that is consistent with all of this evidence.
We develop a life-cycle model in which the household has nonhomothetic utility over two
types of consumption goods, basic and luxury. As the household’s wealth grows, it consumes
relatively more of the luxury good, which has lower curvature than the basic good. The
household’s relative risk aversion falls in wealth, which has implications for consumption and
portfolio choice that are more consistent than those of the power utility model. We calibrate
the model using a standard process for labor income and find that the nonhomothetic model
can account for most of the observed cross-sectional variation in consumption and portfolio
behavior.
More broadly, this paper shows that household portfolios respond to wealth-varying risk
aversion. This finding provides microeconomic foundations for recent representative-agent
asset pricing models that are based on preferences that generate wealth-varying risk aversion
(Aıt-Sahalia, Parker, and Yogo 2004, Campbell and Cochrane 1999).
28
Appendix A Consumer Expenditure Survey
In the Interview Survey component of the CEX, the Bureau of Labor Statistics (BLS) collects
data on the characteristics and major expenditures of U.S. households. The CEX is a rotat-
ing panel that samples approximately 5,000 households every calendar quarter (over 7,000
households more recently). The BLS interviews each household up to five times every three
months before replacement. The first interview is for practice, so only the second through
fifth interviews are available in the public-use data files. The BLS interviews approximately
the same number of households in each of the three months of a calendar quarter, and at each
interview, households report their expenditures from the previous three months. Therefore,
the CEX can be thought of as three non-overlapping panels of quarterly expenditure data.
The BLS estimates that approximately 90–95% of total household expenditures are covered
by the survey.
Although CEX data are available in the present format since 1980, we use the CEX
files from 1982–2002. We do not use data from 1980–1981 since the expenditure on food
was collected with a different questionnaire. The BLS has changed the sampling design
of the CEX on two occasions, between 1985–1986 and between 1995–1996. Consequently,
households cannot be linked across files during these years. Therefore, households in the
1985:4 file are linked to the same households in the early release of 1986:1 data from the
1985 CEX files. Similarly, we use the early release of 1996:1 data from the 1995 CEX files.
We use the Member Characteristics and Income File to identify the reference person of
each household and, for married households, the spouse. We define the household head as
the husband for married households and as the reference person otherwise. Only house-
holds whose head is between ages 26 and 75 at the time of interview are kept for analysis.
Households are grouped by birth cohort and education according to the characteristics of
the household head. We create thirteen birth cohorts in five year bins, from those born
1910–1914 to those born 1970–1974. The four education groups are some high school, high
school graduate, some college, and college graduate.
29
We follow the procedure in Attanasio and Weber (1995) to prepare the CEX files for
empirical analysis. To summarize, we first drop households that live in rural areas, live in
student housing, or are incomplete income respondents. Rural households are dropped since
the BLS failed to survey them during 1981:3–1983:4. Household consumption is constructed
from the Monthly Expenditures File and is defined as all expenditures excluding consumer
durables (as defined in the national accounts), housing, health, and education. Nominal
expenditures are deflated to real 1997 dollars using the Consumer Price Index (CPI) for
all urban consumers. Each expenditure item in the CEX is carefully matched to a region-
and item-specific CPI, so that the price deflator is household specific. To eliminate obvious
data errors, we drop households that report no food or only food expenditures. Monthly
expenditures are summed over all three months of an interview period, which results in total
household consumption for that quarter.
In the second and fifth interviews, the BLS collects income data for the previous twelve
months. Disposable income is computed as total household income after taxes less capital
income and pension contributions. Capital income includes interest on savings accounts and
bonds as well as income from dividends, royalties, estates, and trusts. Pension contribu-
tions is the sum of income contributed to social security, railroad retirement, government
retirement, private pension, and individual retirement plans.
In the fifth interview, the BLS collects financial data for the previous twelve months.
In particular, households report the estimated value of securities such as stocks, mutual
funds, private bonds, and Treasury notes. Following Vissing-Jørgensen (2002), a household
is identified as a stockholder if its holding in these securities was positive twelve months prior
to the interview or has increased in the previous twelve months.
30
Appendix B Survey of Consumer Finances
The SCF is a triennial survey conducted by the Federal Reserve Board. The SCF employs
a dual sampling methodology, combining data collected from a representative sample of ap-
proximately 3,000 U.S. households with data collected from a sample of approximately 1,500
high-wealth households identified through tax-return data. Sample weights are constructed
to adjust for biases caused by missing responses and to create a sample that is representa-
tive of the population as a whole. Aizcorbe, Kennickell, and Moore (2003) and references
therein describe the sampling methodology in detail. We use data that is publicly available
through http://www.federalreserve.gov/Pubs/oss/oss2/scfindex.html. The variable names
below refer to those from the SCF codebook.
We classify survey respondents using the age-class variable (AGECL). This variable di-
vides the population into six age groups (26–35, 36–45, . . . , 75+), based on the age of the
household head. Because we do not model households in the 75+ age group, we focus on
the first five age groups. Within each age group, we weight the observations using the
weight variable (WGT). Using these weights, we construct quartiles for total financial assets
(FIN). We consider five age groups and four asset quartiles for a total of 20 bins. We use
the education-class variable (EDCL) to identify the education level of the household head.
EDCL is equal to 1 for no high school diploma or GED, 2 for high school diploma or GED,
3 for some college, and 4 for a college degree.
Total financial assets includes liquid financial assets (transaction accounts), certificates
of deposit, mutual funds, directly held stocks and bonds, IRAs, thrift accounts and future
pensions, savings bonds, the cash value of life insurance, and “other” managed and financial
assets (loans, future proceeds, royalties, futures, non-public stock, deferred compensation,
oil/gas/mineral investments, cash, trusts, annuities, and managed investment accounts in
which the household has an equity interest). We use the total equity variable (EQUITY)
to determine the household’s equity holdings, which includes both directly-held stock and
stocks held through mutual funds, retirement accounts, trusts, and other managed assets.
31
The portfolio share is equal to EQUITY divided by FIN. The variable HEQUITY is equal to
one if EQUITY is greater than zero and is equal to zero otherwise. For results conditional
on holding stocks, we repeat the analysis on the sub-sample of respondents with HEQUITY
equal to one.
Appendix C Numerical Solution of the Life-Cycle Model
Following the usual methodology, the model is solved backward from the last period of life.
Because it is optimal to consume all of cash-on-hand in the last period, optimal consumption
policy is given by the equations
BT =WT
1 + α1/φBγ/φ−1T
, (20)
LT =WT
1 + α−1/γLφ/γ−1T
. (21)
The value function in period T − 1 is given by
JT−1(WT−1, PT−1) = maxBT−1,LT−1,aT−1
U(BT−1, LT−1) + βET−1[U(BT , LT )]. (22)
We normalize the consumption policy variables by permanent income as bt = Bt/Pt
and lt = Lt/Pt. Similarly, we normalize cash-on-hand as wt = Wt/Pt and savings as st =
wt − bt − lt. Finally, we define the recursive function
jt(wt, Pt) = maxbt,lt,at
b1−γt
1 − γ+ αP γ−φ
t
l1−φt
1 − φ+ βEt[(Gt+1ηt+1)
1−γjt+1(wt+1, Pt+1)]
. (23)
The value function in period T − 1 can be rewritten as
JT−1(WT−1, PT−1) = P 1−γT−1jT−1(wT−1, PT−1). (24)
32
By induction, the value function in any period t is given by
Jt(Wt, Pt) = P 1−γt jt(wt, Pt). (25)
We redefine the life-cycle problem as the solution to Bellman equation (23) subject to
the intertemporal budget constraint,
wt+1 =Rt+1st
Gt+1ηt+1
+ εt+1, (26)
and the law of motion for permanent income (14). The first-order conditions for the Bellman
equation, together with the envelope theorem, imply that
b−γt = Et[βRt+1(Gt+1ηt+1bt+1)
−γ], (27)
0 = Et[βst(Re,t+1 − Rf )(Gt+1ηt+1bt+1)−γ]. (28)
The life-cycle problem is essentially solved through recursion on these equations.
We discretize the joint probability distribution for stock returns and income shocks as
(νi, pνi )I
i=1 = (ν1, pν1), . . . , (νI , p
νI ),
(ηj, pηj )J
j=1 = (η1, pη1), . . . , (ηJ , pη
J),
(εk, pεk)K
k=1 = (ε1, pε1), . . . , (εK , pε
K).
We discretize the state space as
slLl=1 = s1, . . . , sL,
wmMm=1 = w1, . . . , wM,
PnNn=1 = P1, . . . , PN.
33
In each period t, we define the functions
Θt(sl, Pn) =
[I∑
i=1
J∑j=1
K∑k=1
pνi p
ηjp
εkβ[Rf + at(sl, Pn)(Reνi − Rf )](Gt+1ηj)
−γ×
bt+1 (wt+1(sl, Pn; νi, ηj, εk), Gt+1Pnηj)−γ]−1/γ
, (29)
Ωt(sl, Pn) =I∑
i=1
J∑j=1
K∑k=1
pνi p
ηjp
εkβsl(Reνi − Rf )(Gt+1ηj)
−γ ×
bt+1 (wt+1(sl, Pn; νi, ηj, εk), Gt+1Pnηj)−γ , (30)
Γt(sl, Pn) = α1/φP γ/φ−1n Θt(sl, Pn)γ/φ, (31)
where
wt+1(sl, Pn; νi, ηj, εk) =[Rf + at(sl, Pn)(Reνi − Rf )]sl
Gt+1ηj
+ εk. (32)
Starting with the known solution in period T , the life-cycle problem is solved recursively
through the following algorithm.
1. For each point (sl, Pn) on the grid, find at(sl, Pn) such that Ωt(sl, Pn) = 0. If an
interior solution does not exist, at(sl, Pn) = 0 if Ωt(sl, Pn) < 0, and at(sl, Pn) = 1 if
Ωt(sl, Pn) > 0.
2. For each point (sl, Pn) on the grid, compute Θt(sl, Pn) and Γt(sl, Pn).
3. Define wl = sl + Θt(sl, Pn) + Γt(sl, Pn), at(wl, Pn) = at(sl, Pn), and bt(wl, Pn) =
Θt(sl, Pn).
4. For each point (wm, Pn) on the grid, compute at(wm, Pn) by interpolating at(wl, Pn) as
a function of wl and imposing the constraint at(wm, Pn) ∈ [0, 1].
5. For each point (wm, Pn) on the grid, compute bt(wm, Pn) by interpolating bt(wl, Pn) as
a function of wl.
6. Compute lt(wm, Pn) = α1/φPγ/φ−1n bt(wm, Pn)γ/φ.
34
References
Aıt-Sahalia, Yacine, Jonathan A. Parker, and Motohiro Yogo, 2004, Luxury goods and the
equity premium, Journal of Finance 59, 2959–3004.
Aizcorbe, Ana M., Arthur B. Kennickell, and Kevin B. Moore, 2003, Recent changes in U.S.
family finances: Evidence from the 1998 and 2001 Survey of Consumer Finances, Federal
Reserve Bulletin 89, 1–32.
Ameriks, John, and Stephen P. Zeldes, 2004, How do household portfolio shares vary with
age?, Working paper, Columbia University.
Attanasio, Orazio P., and Guglielmo Weber, 1995, Is consumption growth consistent with
intertemporal optimization? Evidence from the Consumer Expenditure Survey, Journal
of Political Economy 103, 1121–1157.
Balduzzi, Pierluigi, and Anthony W. Lynch, 1999, Transaction costs and predictability:
Some utility cost calculations, Journal of Financial Economics 52, 47–78.
Barberis, Nicholas, 2000, Investing for the long run when returns are predictable, Journal of
Finance 55, 225–264.
Benzoni, Luca, Pierre Collin-Dufresne, and Robert S. Goldstein, 2007, Portfolio choice over
the life-cycle when the stock and labor markets are cointegrated, Journal of Finance,
forthcoming.
Bertaut, Carol C., and Michael Haliassos, 1997, Precautionary portfolio behavior from a
life-cycle perspective, Journal of Economic Dynamics and Control 21, 1511–1542.
Bertaut, Carol C., and Martha Starr-McCluer, 2002, Household portfolios in the United
States, in Luigi Guiso, Michael Haliassos, and Tullio Jappelli, ed.: Household Portfolios .
chap. 5, pp. 181–217 (MIT Press: Cambridge, MA).
35
Blume, Marshall E., and Irwin Friend, 1975, The asset structure of individual portfolios and
some implications for utility functions, Journal of Finance 30, 585–603.
Bodie, Zvi, Robert C. Merton, and William F. Samuelson, 1992, Labor supply flexibility
and portfolio choice in a life cycle model, Journal of Economic Dynamics and Control 16,
427–449.
Browning, Martin, and Thomas F. Crossley, 2000, Luxuries are easier to postpone: A proof,
Journal of Political Economy 108, 1022–1026.
Brunnermeier, Markus K., and Stefan Nagel, 2005, Do wealth fluctuations generate time-
varying risk aversion? Micro-evidence on individuals’ asset allocation, Working paper,
Princeton University.
Calvet, Laurent E., John Y. Campbell, and Paolo Sodini, 2006, Down or out: Assessing the
welfare costs of household investment mistakes, NBER Working Paper No. 12030.
Campbell, John Y., 2006, Household finance, Journal of Finance 61, 1553–1604.
Campbell, John Y., Joao F. Cocco, Francisco J. Gomes, and Pascal J. Maenhout, 2001, In-
vesting retirement wealth: A life cycle model, in John Y. Campbell, and Martin Feldstein,
ed.: Risk Aspects of Investment-Based Social Security Reform . chap. 11, pp. 439–473
(University of Chicago Press: Chicago).
Campbell, John Y., and John H. Cochrane, 1999, By force of habit: A consumption-based
explanation of aggregate stock market behavior, Journal of Political Economy 107, 205–
251.
Campbell, John Y., and Luis M. Viceira, 1999, Consumption and portfolio decisions when
expected returns are time-varying, Quarterly Journal of Economics 114, 433–495.
Carroll, Christopher D., 1992, The buffer-stock theory of saving: Some macroeconomic
evidence, Brookings Papers on Economic Activity 2, 61–156.
36
Carroll, Christopher D., 1997, Buffer-stock saving and the life cycle/permanent income hy-
pothesis, Quarterly Journal of Economics 112, 1–55.
Carroll, Christopher D., 2000, Why do the rich save so much?, in Joel B. Slemrod, ed.: Does
Atlas Shrug? The Economic Consequences of Taxing the Rich (Harvard University Press:
Cambridge, MA).
Carroll, Christopher D., 2002, Portfolios of the rich, in Luigi Guiso, Michael Haliassos, and
Tullio Jappelli, ed.: Household Portfolios . chap. 10, pp. 389–429 (MIT Press: Cambridge,
MA).
Carroll, Christopher D., and Andrew A. Samwick, 1997, The nature of precautionary wealth,
Journal of Monetary Economics 40, 41–71.
Cocco, Joao F., 2005, Portfolio choice in the presence of housing, Review of Financial Studies
18, 535–567.
Cocco, Joao F., Francisco J. Gomes, and Pascal J. Maenhout, 2005, Consumption and
portfolio choice over the life cycle, Review of Financial Studies 18, 491–533.
Davis, Steven J., and Paul Willen, 2002, Income shocks, asset returns, and portfolio choice, in
Zvi Bodie, Olivia S. Mitchell, P. Brett Hammond, and Stephen Zeldes, ed.: Innovations in
Retirement Financing . chap. 2, pp. 20–49 (University of Pennsylvania Press: Philadelphia,
PA).
Friend, Irwin, and Marshall E. Blume, 1975, The demand for risky assets, American Eco-
nomic Review 65, 900–922.
Gakidis, Harry E., 1998, Stocks for the old? Earnings uncertainty and life-cycle portfolio
choice, Ph.D. dissertation, MIT.
37
Gomes, Francisco, and Alexander Michaelides, 2003, Portfolio choice with internal habit
formation: A life-cycle model with uninsurable labor income risk, Review of Economic
Dynamics 6, 729–766.
Gomes, Francisco, and Alexander Michaelides, 2005, Optimal life-cycle asset allocation: Un-
derstanding the empirical evidence, Journal of Finance 60, 869–904.
Gourinchas, Pierre-Olivier, and Jonathan A. Parker, 2002, Consumption over the life cycle,
Econometrica 70, 47–89.
Guiso, Luigi, Michael Haliassos, and Tullio Jappelli, 2002, Household Portfolios (MIT Press:
Cambridge, MA).
Haliassos, Michael, and Carol C. Bertaut, 1995, Why do so few hold stocks?, Economic
Journal 105, 1110–1129.
Hanoch, Giora, 1977, Risk aversion and consumer preferences, Econometrica 45, 413–426.
Heaton, John, and Deborah Lucas, 1997, Market frictions, savings behavior, and portfolio
choice, Macroeconomic Dynamics 1, 76–101.
Heaton, John, and Deborah Lucas, 2000, Portfolio choice and asset prices: The importance
of entrepreneurial risk, Journal of Finance 55, 1163–1198.
Hu, Xiaoqing, 2005, Portfolio choices for homeowners, Journal of Urban Economics 58,
114–136.
Kim, Tong Suk, and Edward Omberg, 1996, Dynamic nonmyopic portfolio behavior, Review
of Financial Studies 9, 141–161.
Lynch, Anthony W., and Sinan Tan, 2006, Labor income dynamics and business-cycle fre-
quencies: Implications for portfolio choice, Working paper, New York University.
38
Mankiw, N. Gregory, and Stephen P. Zeldes, 1991, The consumption of stockholders and
nonstockholders, Journal of Financial Economics 29, 97–112.
Polkovnichenko, Valery, 2007, Life-cycle portfolio choice with additive habit formation pref-
erences and uninsurable labor income risk, Review of Financial Studies 20, 83–124.
Roussanov, Nikolai, 2006, Diversification and its discontents: Idiosyncratic and en-
trepreneurial risk in the quest for social status, Working paper, University of Chicago.
Samuelson, Paul A., 1969, Lifetime portfolio selection by dynamic stochastic programming,
Review of Economics and Statistics 51, 239–246.
Stiglitz, Joseph E., 1969, Behavior towards risk with many commodities, Econometrica 4,
660–667.
Storesletten, Kjetil, Christopher I. Telmer, and Amir Yaron, 2007, Asset pricing with idiosyn-
cratic risk and overlapping generations, Review of Economic Dynamics, forthcoming.
Viceira, Luis M., 2001, Optimal portfolio choice for long-horizon investors with nontradable
labor income, Journal of Finance 56, 433–470.
Vissing-Jørgensen, Annette, 2002, Limited asset market participation and the elasticity of
intertemporal substitution, Journal of Political Economy 110, 825–853.
Wachter, Jessica A., 2002, Portfolio and consumption decisions under mean-reverting re-
turns: An exact solution for complete markets, Journal of Financial and Quantitative
Analysis 37, 63–91.
Wachter, Jessica A., and Motohiro Yogo, 2007, Supplementary material for “Why do house-
hold portfolio shares rise in wealth?”, Working paper, University of Pennsylvania.
Yao, Rui, and Harold H. Zhang, 2005, Optimal consumption and portfolio choices with risky
housing and borrowing constraints, Review of Financial Studies 18, 197–239.
39
Yogo, Motohiro, 2006, Permanent income and consumption volatility, Working paper, Uni-
versity of Pennsylvania.
Zeldes, Stephen P., 1989, Optimal consumption with stochastic income: Deviations from
certainty equivalence, Quarterly Journal of Economics 104, 275–298.
40
Table 1: Basic Expenditure Share and Consumption Volatility by Age and Total Consump-tionWe sort the sample of households in the 1982–2002 CEX into age (columns) groups, thenquartiles of nondurable and service consumption (rows) within each age group. Panel Areports the median (within each bin) of total consumption. Panel B reports the median(within each bin) of basic consumption as a share of total consumption. Panel C reportsthe median (within each bin) of consumption volatility in units of percentage growth perquarter. For each household, consumption volatility is measured as the mean squared logconsumption growth over two quarters minus that over one quarter.
Percentile of Consumption Age26–35 36–45 46–55 56–65 66–75
Panel A: Consumption (Thousands of 2001 Dollars)0–25 3.4 3.7 3.6 3.4 3.225–50 5.1 5.5 5.5 5.2 4.950–75 6.8 7.4 7.5 7.1 6.975–100 10.0 10.8 11.3 11.1 11.0Top 5 15.5 16.4 17.8 18.8 19.3All Households 5.8 6.4 6.5 6.1 5.8
Panel B: Basic Consumption (% of Consumption)0–25 64.2 64.0 64.7 68.1 72.625–50 56.7 55.4 54.9 58.5 62.350–75 50.8 50.0 48.3 51.0 56.075–100 42.2 42.1 39.7 41.7 44.8Top 5 32.0 35.4 31.6 31.1 34.3All Households 53.9 52.8 51.7 54.9 58.7
Panel C: Standard Deviation of Consumption Growth0–25 9.2 9.2 9.3 10.2 8.725–50 8.2 6.4 6.9 8.8 6.150–75 8.9 7.1 7.5 7.5 8.775–100 10.6 10.2 11.9 11.4 10.5Top 5 15.2 15.8 16.8 15.8 15.2All Households 9.2 8.2 8.7 9.4 8.4
41
Table 2: Nondurable Expenditure Share by Age and Total ConsumptionWe sort the sample of households in the 1982–2002 CEX into age (columns) groups, thenquartiles of nondurable and service consumption (rows) within each age group. The levels oftotal consumption that define the quartiles are reported in Panel A of Table 1. The panelsreport the median (within each bin) of the components of nondurable consumption as ashare of total consumption.
Percentile of Consumption Age26–35 36–45 46–55 56–65 66–75
Panel A: Food at Home (% of Consumption)0–25 31.4 31.2 30.9 31.3 35.125–50 24.8 25.4 24.5 25.8 28.450–75 20.7 22.5 21.2 22.0 24.075–100 16.6 17.3 16.2 16.8 17.9Top 5 12.1 12.7 11.7 11.5 12.0All Households 22.7 23.6 22.5 23.7 25.9
Panel B: Food Away from Home (% of Consumption)0–25 4.7 4.5 4.0 2.8 1.725–50 6.7 7.2 7.2 5.6 4.850–75 8.5 8.4 8.5 7.4 6.175–100 10.9 9.7 10.7 9.7 8.8Top 5 13.7 11.4 11.2 11.5 10.3All Households 7.5 7.4 7.5 6.3 5.3
Panel C: Clothing and Shoes (% of Consumption)0–25 5.1 4.3 2.8 1.8 0.825–50 6.2 6.1 5.3 4.3 2.750–75 6.7 7.2 6.5 5.3 4.275–100 6.8 7.6 7.0 6.1 5.1Top 5 6.7 7.8 7.0 6.0 4.5All Households 6.2 6.3 5.4 4.4 3.2
Panel D: Fuel Oil, Coal, and Gasoline (% of Consumption)0–25 8.9 8.6 8.4 7.9 5.725–50 8.5 8.0 8.5 8.0 6.650–75 7.5 7.3 7.7 7.5 6.475–100 5.8 5.6 5.9 5.7 4.9Top 5 4.2 3.9 4.4 4.2 3.5All Households 7.6 7.3 7.5 7.2 5.9
Panel E: Other Nondurable Goods (% of Consumption)0–25 3.6 3.5 3.3 3.1 2.425–50 4.1 4.0 3.9 4.3 3.450–75 4.2 4.2 4.0 4.3 3.875–100 3.7 3.7 3.3 3.9 3.3Top 5 3.1 3.3 2.8 2.9 2.4All Households 3.9 3.9 3.6 3.9 3.2
42
Table 3: Service Expenditure Share by Age and Total ConsumptionWe sort the sample of households in the 1982–2002 CEX into age (columns) groups, thenquartiles of nondurable and service consumption (rows) within each age group. The levels oftotal consumption that define the quartiles are reported in Panel A of Table 1. The panelsreport the median (within each bin) of the components of service consumption as a share oftotal consumption.
Percentile of Consumption Age26–35 36–45 46–55 56–65 66–75
Panel A: Household Operation (% of Consumption)0–25 20.6 20.5 21.2 23.8 26.025–50 19.8 18.6 18.9 20.7 23.250–75 18.6 17.0 16.5 17.6 20.875–100 15.7 15.5 14.4 14.7 17.0Top 5 12.2 14.4 12.8 12.0 13.7All Households 18.8 17.9 17.6 18.9 21.5
Panel B: Transportation (% of Consumption)0–25 4.5 5.5 5.7 4.6 2.225–50 7.5 8.0 9.2 8.0 6.650–75 8.8 8.9 10.4 10.2 9.075–100 10.8 10.6 12.5 11.3 11.1Top 5 12.9 11.1 13.8 11.6 11.1All Households 8.0 8.2 9.5 8.8 7.2
Panel C: Personal Care (% of Consumption)0–25 2.0 1.9 1.8 1.7 1.925–50 2.1 2.0 2.0 2.1 2.250–75 2.1 2.1 2.0 2.1 2.375–100 2.2 2.1 2.2 2.2 2.1Top 5 2.2 2.1 2.0 1.9 1.6All Households 2.1 2.0 2.0 2.0 2.1
Panel D: Personal Business (% of Consumption)0–25 1.3 2.0 2.2 2.0 0.625–50 3.2 4.0 3.8 3.2 1.950–75 3.5 4.5 4.5 4.0 2.575–100 3.8 4.4 4.8 4.3 2.9Top 5 3.5 4.7 5.5 5.0 4.3All Households 3.0 3.8 3.9 3.4 2.0
Panel E: Recreation (% of Consumption)0–25 2.7 3.2 2.9 2.8 1.725–50 4.0 4.4 4.0 3.4 3.250–75 4.7 4.9 4.8 3.8 3.775–100 5.5 5.9 5.8 4.8 4.2Top 5 6.8 6.3 6.2 6.0 5.3All Households 4.2 4.6 4.3 3.7 3.4
43
Table 4: Portfolio Share by Age and WealthWe sort the sample of households in the 2001 SCF into age (columns) groups, then quartilesof financial wealth (rows) within each age group. Panel A reports the median (within eachbin) of financial wealth. Panel B reports the percentage of households that own stocks withineach bin. Panel C reports the median (within each bin) of the portfolio share in stocks.
Percentile of Financial Assets Age26–35 36–45 46–55 56–65 66–75
Panel A: Financial Assets (Thousands of 2001 Dollars)0–25 0.0 0.4 0.5 0.4 0.625–50 1.9 9.0 19.3 18.6 17.350–75 9.4 44.7 80.8 106.8 100.675–100 60.8 205.8 385.0 558.9 525.1Top 5 262.0 658.4 1343.3 2655.0 2293.0All Households 4.2 22.1 40.0 47.1 42.5
Panel B: Percentage of Households with Stocks0–25 3 9 9 5 125–50 36 52 53 46 1550–75 71 83 78 83 5375–100 86 93 96 95 88Top 5 91 97 98 99 96All Households 49 60 59 57 39
Panel C: Stocks as Percentage of Financial Assets0–25 0 0 0 0 025–50 0 6 8 0 050–75 31 47 36 38 175–100 45 59 59 54 60Top 5 60 71 63 67 59All Households 7 29 25 23 0
44
Table 5: Portfolio Share by Age and Wealth for StockholdersWe sort the sub-sample of stockholding households in the 2001 SCF into age (columns)groups, then quartiles of financial wealth (rows) within each age group. Panel A reports themedian (within each bin) of financial assets. Panel B reports the median (within each bin)of the portfolio share in stocks.
Percentile of Financial Assets Age26–35 36–45 46–55 56–65 66–75
Panel A: Financial Assets (Thousands of 2001 Dollars)0–25 2.3 9.1 18.2 28.0 51.425–50 10.0 39.6 68.0 105.3 190.550–75 32.6 94.0 185.0 269.9 409.675–100 119.2 310.4 609.0 989.2 1079.0Top 5 459.3 850.0 2009.1 3836.3 4005.0All Households 18.0 62.8 105.3 161.4 269.5
Panel B: Stocks as Percentage of Financial Assets0–25 45 43 39 45 3925–50 54 49 47 48 4750–75 48 56 49 53 6475–100 60 68 67 54 62Top 5 67 69 69 55 58All Households 49 53 49 50 58
45
Table 6: Portfolio Share by Education, Age, and Wealth for StockholdersWe sort the sub-sample of stockholding households in the 2001 SCF into education (pan-els) and age (columns) groups, then quartiles of financial wealth (rows) within each educa-tion/age group. The levels of financial wealth that define the quartiles (not reported) areeducation/age specific. Each panel reports the median (within each bin) of the portfolioshare in stocks.
Percentile of Financial Assets Age26–35 36–45 46–55 56–65 66–75
Panel A: High School Graduates0–25 50 41 47 31 3925–50 47 49 60 39 950–75 49 57 59 68 3775–100 31 58 60 46 62Top 5 11 47 81 47 51All Households 42 50 50 43 41
Panel B: Some College0–25 43 38 32 47 3625–50 48 46 38 29 6850–75 50 47 47 41 4275–100 54 55 50 46 57Top 5 67 88 46 38 71All Households 48 48 42 46 47
Panel C: College Graduates0–25 45 44 41 45 5625–50 41 55 46 53 6450–75 53 69 60 66 8075–100 68 71 70 64 64Top 5 62 59 74 55 64All Households 54 57 55 58 70
46
Tab
le7:
Par
amet
ers
inth
eB
ench
mar
kC
alib
rati
onT
he
table
repor
tspar
amet
ers
use
din
the
ben
chm
ark
calibra
tion
ofth
elife
-cycl
eco
nsu
mpti
onan
dpor
tfol
io-c
hoi
cem
odel
wit
hst
och
asti
cla
bor
inco
me.
Inth
epow
eruti
lity
model
,th
ehou
sehol
dhas
pow
eruti
lity
over
one
consu
mpti
ongo
od.
Inth
enon
hom
othet
icm
odel
,th
ehou
sehol
dhas
addit
ivel
yse
par
able
and
non
hom
othet
icuti
lity
over
two
consu
mpti
ongo
ods,
abas
ican
da
luxury
good.
The
model
sar
esi
mula
ted
atan
annual
freq
uen
cy.
Par
amet
erSym
bol
Non
hom
othet
icPow
erP
refe
rence
s:D
isco
unt
Fac
tor
β0.
960.
96C
urv
ature
for
Bas
icG
ood
γ7
5C
urv
ature
for
Luxury
Good
φ3
Uti
lity
Wei
ght
onLuxury
Good
α10
0Lab
orIn
com
e:SD
ofPer
man
ent
Inco
me
ση
15%
15%
SD
ofTra
nsi
tory
Inco
me
σε
21%
21%
Ass
etR
eturn
s:R
iskfr
eeR
ate
Rf−
12%
2%E
quity
Pre
miu
mR
e−
Rf
4%4%
SD
ofSto
ckR
eturn
sσ
ν18
%18
%C
orre
lati
onbet
wee
nPer
man
ent
Inco
me
and
Ret
urn
sρ
15%
15%
47
Table 8: Basic Expenditure Share in the Nonhomothetic ModelPanel A reports the median of basic consumption as a share of total consumption for stock-holding households, tabulated with the 1982–2002 CEX. Panel B reports the median of basicexpenditure share for households simulated in the nonhomothetic model.
Percentile of Consumption Age26–35 36–45 46–55 56–65 66–75
Panel A: CEX (Stockholders Only)0–25 56 56 55 56 6325–50 50 50 48 51 5450–75 46 45 43 46 4875–100 38 38 35 36 39Top 5 28 31 28 26 29All Households 48 47 45 47 51
Panel B: Nonhomothetic Utility0–25 82 73 70 68 6725–50 68 59 55 53 5250–75 58 49 45 42 4175–100 46 38 33 30 28Top 5 38 30 26 23 22All Households 63 54 50 47 46
48
Table 9: Consumption Volatility in the Life-Cycle ModelPanel A reports the median of consumption volatility for stockholding households, tabulatedwith the 1982–2002 CEX. Panel B (Panel C) reports the median of consumption volatilityfor households simulated in the nonhomothetic (power utility) model.
Percentile of Consumption Age26–35 36–45 46–55 56–65 66–75
Panel A: CEX (Stockholders Only)0–25 5.8 7.6 8.9 6.5 8.025–50 7.9 7.3 8.4 6.0 4.250–75 8.2 7.0 8.1 7.8 7.275–100 11.6 10.6 14.5 12.4 7.8Top 5 15.7 15.3 19.1 24.8 15.6All Households 8.2 8.0 9.3 8.3 7.0
Panel B: Nonhomothetic Utility0–25 16.1 7.6 6.1 4.9 2.925–50 10.1 8.1 6.7 5.5 3.450–75 10.0 8.4 7.1 6.0 3.775–100 10.1 9.0 7.9 6.6 3.9Top 5 9.7 8.9 8.9 6.7 4.1All Households 11.0 8.3 6.9 5.8 3.4
Panel C: Power Utility0–25 15.0 7.6 6.2 5.0 2.825–50 9.6 7.7 6.4 5.2 2.850–75 9.3 7.8 6.5 5.5 2.875–100 9.1 8.0 6.7 5.7 2.8Top 5 8.8 7.9 6.7 6.0 2.8All Households 10.2 7.8 6.5 5.3 2.8
49
Table 10: Portfolio Share in the Life-Cycle ModelPanel A reports the median of portfolio share in stocks for stockholding households, tabulatedwith the 2001 SCF (also reported in Panel B of Table 5). Panel B (Panel C) reports themedian of portfolio share for households simulated in the nonhomothetic (power utility)model.
Percentile of Financial Assets Age26–35 36–45 46–55 56–65 66–75
Panel A: 2001 SCF (Stockholders Only)0–25 45 43 39 45 3925–50 54 49 47 48 4750–75 48 56 49 53 6475–100 60 68 67 54 62Top 5 67 69 69 55 58All Households 49 53 49 50 58
Panel B: Nonhomothetic Utility0–25 100 68 46 40 4225–50 100 73 54 47 4550–75 100 77 59 52 4875–100 100 79 64 58 53Top 5 100 82 67 59 55All Households 100 74 56 49 47
Panel C: Power Utility0–25 100 84 51 41 4025–50 100 73 50 41 3950–75 100 68 49 41 3975–100 99 62 48 41 39Top 5 93 61 47 41 38All Households 100 71 49 41 39
50
Table 11: Portfolio Share in the Life-Cycle Model with Unemployment RiskPanel A reports the median of portfolio share in stocks for stockholding households, tabulatedwith the 2001 SCF (also reported in Panel B of Table 5). Panel B (Panel C) reports themedian of portfolio share for households simulated in the nonhomothetic (power utility)model. The probability of unemployment in the model is 0.5%.
Percentile of Financial Assets Age26–35 36–45 46–55 56–65 66–75
Panel A: 2001 SCF (Stockholders Only)0–25 45 43 39 45 3925–50 54 49 47 48 4750–75 48 56 49 53 6475–100 60 68 67 54 62Top 5 67 69 69 55 58All Households 49 53 49 50 58
Panel B: Nonhomothetic Utility0–25 38 66 46 40 4225–50 100 71 53 47 4550–75 100 76 59 52 4875–100 100 78 64 58 53Top 5 100 80 65 59 53All Households 99 73 55 49 47
Panel C: Power Utility0–25 51 82 51 41 3925–50 100 72 49 41 3950–75 100 67 48 41 3875–100 98 61 48 41 38Top 5 92 60 47 42 38All Households 100 70 49 41 39
51
0
0.5
1
1.5
2
0
2
4
6
8
10
0
0.2
0.4
0.6
0.81
Per
m In
com
e
Bas
ic C
onsu
mpt
ion
at A
ge 5
0
Cas
h−on−
Han
d
Fraction of PI
0
0.5
1
1.5
2
0
2
4
6
8
10
0
0.2
0.4
0.6
0.81
Per
m In
com
e
Luxu
ry C
onsu
mpt
ion
at A
ge 5
0
Cas
h−on−
Han
d
Fraction of PI
Fig
ure
1:O
pti
mal
Con
sum
pti
onPol
icy
inth
eN
onhom
othet
icM
odel
The
figu
resh
ows
the
opti
mal
consu
mpti
onpol
icy
atag
e50
for
alife
-cycl
eco
nsu
mpti
onan
dpor
tfol
io-c
hoi
cem
odel
wit
hnon
hom
othet
icuti
lity
.T
he
hou
sehol
dre
ceiv
esst
och
asti
cla
bor
inco
me
from
ages
26th
rough
65an
ddie
sat
age
75.
The
stat
eva
riab
les
are
nor
mal
ized
cash
-on-h
and
(w=
W/P
)an
dper
man
ent
inco
me
(P).
The
pol
icy
vari
able
sar
ebas
icco
nsu
mpti
on(B
)an
dlu
xury
consu
mpti
on(L
),bot
hsh
own
asa
frac
tion
ofper
man
ent
inco
me.
Tab
le7
repor
tsth
ein
com
ean
dpre
fere
nce
par
amet
ers
ofth
em
odel
.
52
0
0.5
1
1.5
2
0
2
4
6
8
100
0.2
0.4
0.6
0.8
1
Perm Income
Portfolio Share at Age 50
Cash−on−Hand
Sha
re in
Sto
cks
Figure 2: Optimal Portfolio Policy in the Nonhomothetic ModelThe figure shows the optimal portfolio policy at age 50 for a life-cycle consumption andportfolio-choice model with nonhomothetic utility. The household receives stochastic laborincome from ages 26 through 65 and dies at age 75. The state variables are normalizedcash-on-hand (w = W/P ) and permanent income (P ). The policy variable is the portfolioshare in stocks (a). Table 7 reports the income and preference parameters of the model.
53